Since Adleman first proposed DNA computing for the Hamiltonian path problem, several authors have reported DNA computing for 3-SAT. Previous research presented DNA computing on surfaces and demonstrated how to solve a four-variable four-clause instance of 3-SAT, and claimed that the surface-based approach was designed to scale up to larger problems. In this paper we establish an error model for the incomplete "mark" and imperfect "destroy" operations. By using the error model we argue that no matter how large the "mark" and "destroy" rates are we can always give satisfiable instances of 3-SAT such that no DNA strands remain on the surface at the end of the computation. By the surface-based approach the satisfiable instances of 3-SAT would be misdetermined to be unsatisfiable. Thus, the error leads to an incorrect result of the SAT computation. Furthermore, given the "mark" rate p and the "not-destroy" rate rho, we find that the approach can only solve at most N-variable instances of 3-SAT problems, where N=[(2+beta(2)+2+2 square root beta (2))/beta(2)] in which beta=1-1/(p+rhoq) and q=1-p and [a] is the greatest integer less than a or equal to a.